The Research On Static Security Analysis System Of Power System Based On GPU

In the field of power system, static security analysis is a key technology to ensure the stability of power system. Due to the large amount of concurrent computing, the calculation in the classical architecture of server is difficult to ensure real-time capability of the analysis. With the continuous development of GPU technology, CUDA architecture enables GPU in general computing and fulfills the need of large amount of numeration in power flow calculation.Static security analysis is essentially computing power flow for many times. Power flow is essentially solving large sparse nonlinear equation. Normal method to solve nonlinear equation is to iterate, but iteration program does not work well on GPU. So this paper researches on how to reduce iteration and allocate tasks to GPU to be more efficiency. Based on the characteristics of the algorithm of static security analysis and the architecture of GPU hardware, this paper proposed a novel algorithm to deal with small partitioned matrix of sparse matrix and to optimize the operations on small dimensional matrices. In addition, the system can support GPU cluster by importing parallel programming communication interface of high-performance. Comparing with the static security analysis system running on CPU, we decreased calculation time and storage cost on GPU.Experimental results indicate that the proposed algorithm on GPU improves system performance. Our results show up to 70% speedup than GPU with simulation scale from 3000 to 6000. Furthermore, the algorithm is not limited to the application scenarios of static security analysis. It can also be applied to other similar calculation of high performance computing.